Good piece. As I understand it, the “butterfly effect” is an outcome of non linear terms which elevate the noise processes into being an indeterminant factor in the final outcome. While the “political effect” is when dissenting scientists are run out of town on a rail. It’s the latter that’s really destroying scientific integrity and confidence.
When are they useful? In predicting what the weather will be over the next day or two or how the events in one season will play out over the next season. Both useful because of their immediacy, because they can be changed as the future reveals itself, and because judgement can be made of their validity over a time period that allows for validity to be determined and fixes to the model can be tried.
It is super convenient that climate models predict things so far in the future that it is impossible to test their validity but the course of action recommended must be immediately implemented. Faith and hope are innately human but they are both magical thinking.
I’m a former private pilot. Before every flight, I’d check the weather along my route. As I fkew, I’d watch to see if the actual weather was similar to the forecast. Weather forecasters have been known to lie (or more fairly, to be wrng). If a weather forecast over an area of a few hundred miles and a span of a few hours can be wrong (and they sometimes were), I have a hard time believing that a computer mdel can accurately predict the worldwide climate over a period of decades.
I think you’re conflating weather and climate. The difference isn’t so much about distances but time scales. Weather is a high frequency phenomenon changing day to day or morning to evening. Climate change moves much slower.
A person could also be conflating an analogy with the actual thing.
Is there any numerical-methods intuition as to why a large-scale and chaotic system that is not predictable on a short time scale becomes predictable on a much longer time scale?
High frequency random noise superimposed on meaningful low frequency dynamics?
And yet the trends over decades are themselves noise over the longer periods in which climate changes. The future climate has always been uncertain and we have an evolutionary fear of what it might bring.
Just think about how fear of an uncertain future has driven humanity from the way we produce and store food, to water systems, to cultures built on sacrifice to nature gods.
Much of AGW alarmism has nothing to do with science and instead revolves around a false sense of knowledge, divinity of self, preying on primal fears, animistic and nature worship, with a heavy dose of flagellation and sacrifice.
Yes. We can’t predict tomorrow’s weather accurately, but we can for sure predict the weather decades from now. Honest.
You realize that weather forecasts are over 90% correct over 5-day interval ? Also, seasonality is entirely predictable. And that some climate models from 20 years ago have been largely proven out over our lifetime ?
Meteorologists have been studying weather for a very long time. People have been writing weather computer models for decades with constant feedback of real world results to improve their quality. Despite all that, the best forecasts are only accurate for a matter of hours and even then are frequently wrong.
Climate can be thought of as the pural of weather. Climate models aren’t nearly as mature as weather models. From what I’ve read, none of them have been validated against the real world climate and none are doing a very good job of matching actual climatic conditions. Despite all that, we’re supposed to take their word for it, turn over trillions of dollars of wealth while turning over vast power to bureaucrats. Sorry, but no.
Just like any model. When it makes accurate predictions about the actual systems in question ahead of time with minimum effort.
If your model doesn’t match reality, it isn’t reality that’s wrong.
This is a pretty good summary. The problem I see with climate models is they’re attempting to predict very low frequency phenomena, so it’s tough to tell if it’s making accurate predictions.
It’s worse than that – most of the predictions made in their first report have already been falsified.
There are a couple of websites http://www.windy.com and http://www.ventusky.com where you can see nice animations of the weather models in action. A long time ago I was an aviation weather forecaster for a few years and sure wish we had had those then. It is interesting the watch the 7 day prediction and day by day see how the prediction changes for that day as it gets closer.
Paul wrote: “Is there any numerical-methods intuition as to why a large-scale and chaotic system that is not predictable on a short time scale becomes predictable on a much longer time scale?”
EXACTLY THAT. I’d like a formal proof of that please before ANY further climate predictions based on these models are made.
Pug wrote: “High frequency random noise superimposed on meaningful low frequency dynamics?”
How do you know the low frequency stuff isn’t random noise also?
Sure in the longer term there is the physical constraint of the (slightly variable and more so in the ultraviolet) solar energy input but there are physical constraints on the short term weather stuff too.
In any case the models only reflect the assumptions built in to them and as Freeman Dyson has pointed out, ignore biology totally. Biology of course includes things like vegetation cover which changes albedo and far more subtle things like complex iodine compounds released by Krill in the oceans around Antarctica which affect cloud nucleation. In the long term these are important as shown by the current greening of the planet. Can’t figure why that’s a bad thing.
I’m living in Australia, the world’s crash test dummy for “renewable” energy and boy, are we ever dummies. Hundreds of years worth of coal and 40% of the easily recoverable uranium and we have no nukes and are shutting down and blowing up coal plants and putting up with power cuts in major cities because demand exceeds generation capacity. Our renewable sources produced 4% of load during the hot weather the other day.
I don’t know for certain that it’s not noise (I’m not also prescribing radical changes to world economies based on these models), but chemical and geological records indicate multiple slow moving phenomena like ice ages.
The classic test for any physical system model that requires tuning from observation to achieve accuracy is: once the model has been tuned to one set of data, can it reproduce results from another set of data to which it was not tuned?
An example that I’m intimately familiar with: flight dynamics models of aircraft. I have derived a set of differential equations that describe the airplane’s dynamics, given a certain set of assumptions; said equations have parameters that I know a priori – wind tunnel data, structural mode frequencies and shapes, actuator dynamics – but there are certainly errors in my knowledge of the parameters, and there may be errors in the assumptions I used in deriving the equations of motion – e.g., I may have assumed that the aircraft structural modes are not significantly coupled with the aircraft rigid body degrees of freedom. When I go test the aircraft, I compare my model to the truth of the measurements – using a subset of the available data – and adjust the parameters and, if needed, the assumptions that went into my equations of motion, until the model matches the flight results. I then use the updated model to see if I can reproduce the data that I didn’t use in my model-updating process; if I can, then my model has predictive power and is useful for further analyses. If not, then I need to work on my model some more.
This process needs to be applied to the climate models. Take all of the data up through, say, 1960, and adjust a given physical (not correlational) model until it matches. And not just global temperature anomaly either. Going back to my airplane example, if all it predicts accurately is total fuel consumption, it’s not useful; I need it to predict accurately everything from takeoff performance to inner loop stability margins. Similarly for a candidate climate model, it should predict broad trends across each major landmass, ocean surface and depth temperatures across the globe, etc. Now take that model, and run it – with NO additional tweaking – from 1960 to the present day, and see how it does. If the model predicts significant results don’t match reality over that interval – it’s not enough to match sometimes, it has to be a good match across the whole time interval – then it may have predictive power over the next 40+ years and it may behoove us to pay attention to it. If not, it is by definition not a useful model and it is, in the immortal words of John Nance Gardner, “not worth a bucket of warm spit.”
I think there was a different bodily fluid in the bucket in Garner’s original formulation.
Of course another issue is that if you do this process “many times” until the second data set matches as well, you have really just tuned to both data sets and your model still isn’t worth anything.
For the model to be worth something, it must work on the second data set first time. Not on the hundredth try.
That’s basically what we’re dealing with. The climate model projections are essentially extrapolations of an elaborate curve fit. With a rich enough functional basis, one can fit any model to a historical baseline with arbitrary precision. But, that alone does not establish the model as valid. It is necessary, but not sufficient.
Looking over the long term climate record, Earth’s climate exhibits the character of a strange attractor (https://en.wikipedia.org/wiki/Attractor#Strange_attractor) with 2 major modes: A warm mode that we are currently in, and a cool mode. The long term climate prediction challenge is predicting when the mode will change. Shorter term the variations are of smaller magnitude.
I’ve put an embarrassingly large amount of time into studying and thinking about climate models. I certainly resonate with cthulhu’s post, above, especially with: “The classic test for any physical system model that requires tuning from observation to achieve accuracy is: once the model has been tuned to one set of data, can it reproduce results from another set of data to which it was not tuned?”
That test can be used to screen climate models without actually doing the experiment. If a climate model is based in any part on CFD solutions of the Navier-Stokes equations, the answer is a simple and resounding “no.”
When even the best CFD software is used to analyze flow involving turbulence (and turbulence dominates Earth’s hydrodynamics), it must be tuned to the boundary conditions by test. Anything from a new pipe elbow shape to a new aircraft outer moldline constitute new boundary conditions. Tuning CFD to the new boundary conditions (and flow initial conditions) will allow one to make predictions of the results of minor variations in flow initial conditions. But change the boundary conditions, and the tuned CFD won’t produce results even close to what one gets in test.
There are fundamental problems with the Navier-Stokes equations, IMHO, but the universally recognized one is turbulent closure. It isn’t mathematically possible with strictly physical constraints. The constraints used to force closure involve various non-physical mathematical equations which match the number of equations to the number of unknowns plus adjustable parameters. The latter are determined by experiment, and fit only one set of circumstances. It is literally a process of curve-fitting a set of data. Interpolation can be done accurately, but extrapolation puts one in unknown territory.
In short, N-S based climate models are not, and can never be, useful for projecting climate.
Pug,
Ice ages. That. They have exactly zero to do with any human influence as did the temperature variations during the current interglacial before the last couple of hundred years for sure.
I tear my hair out when the media drivel comes on TV. Generally takes less than 2 minutes of watching TV news. Maybe I’m getting old and crotchetty.
Does the USA take climate refugees? Not because of climate change but because of governments becoming totalitarian and oppressive in their “solutions” to the non problem while completely wrecking a country’s economy.
Mike,
I don’t think I did a good job of making my first point clear. I’m aware that ice ages had nothing to do with human intervention – I was listing their presence as an example of low frequency climate dynamics. I think you and I are more or less on the same page.
It is difficult to predict high frequency dynamics because they are so sensitive to modeling and measurement error. E.g., a half microsecond delay in a one microsecond cycle completely inverts the phase.
However, it is also difficult to predict low frequency dynamics because one needs very long timelines to observe them. One can fit a low frequency model that appears to track results over a finite time. But, then one generally finds there is a point at which it starts to diverge, as more of the long term process reveals itself.
When are they useful? When they support the political ambitions of left wing political parties.
Good piece. As I understand it, the “butterfly effect” is an outcome of non linear terms which elevate the noise processes into being an indeterminant factor in the final outcome. While the “political effect” is when dissenting scientists are run out of town on a rail. It’s the latter that’s really destroying scientific integrity and confidence.
When are they useful? In predicting what the weather will be over the next day or two or how the events in one season will play out over the next season. Both useful because of their immediacy, because they can be changed as the future reveals itself, and because judgement can be made of their validity over a time period that allows for validity to be determined and fixes to the model can be tried.
It is super convenient that climate models predict things so far in the future that it is impossible to test their validity but the course of action recommended must be immediately implemented. Faith and hope are innately human but they are both magical thinking.
I’m a former private pilot. Before every flight, I’d check the weather along my route. As I fkew, I’d watch to see if the actual weather was similar to the forecast. Weather forecasters have been known to lie (or more fairly, to be wrng). If a weather forecast over an area of a few hundred miles and a span of a few hours can be wrong (and they sometimes were), I have a hard time believing that a computer mdel can accurately predict the worldwide climate over a period of decades.
I think you’re conflating weather and climate. The difference isn’t so much about distances but time scales. Weather is a high frequency phenomenon changing day to day or morning to evening. Climate change moves much slower.
A person could also be conflating an analogy with the actual thing.
Is there any numerical-methods intuition as to why a large-scale and chaotic system that is not predictable on a short time scale becomes predictable on a much longer time scale?
High frequency random noise superimposed on meaningful low frequency dynamics?
And yet the trends over decades are themselves noise over the longer periods in which climate changes. The future climate has always been uncertain and we have an evolutionary fear of what it might bring.
Just think about how fear of an uncertain future has driven humanity from the way we produce and store food, to water systems, to cultures built on sacrifice to nature gods.
Much of AGW alarmism has nothing to do with science and instead revolves around a false sense of knowledge, divinity of self, preying on primal fears, animistic and nature worship, with a heavy dose of flagellation and sacrifice.
Yes. We can’t predict tomorrow’s weather accurately, but we can for sure predict the weather decades from now. Honest.
You realize that weather forecasts are over 90% correct over 5-day interval ? Also, seasonality is entirely predictable. And that some climate models from 20 years ago have been largely proven out over our lifetime ?
Meteorologists have been studying weather for a very long time. People have been writing weather computer models for decades with constant feedback of real world results to improve their quality. Despite all that, the best forecasts are only accurate for a matter of hours and even then are frequently wrong.
Climate can be thought of as the pural of weather. Climate models aren’t nearly as mature as weather models. From what I’ve read, none of them have been validated against the real world climate and none are doing a very good job of matching actual climatic conditions. Despite all that, we’re supposed to take their word for it, turn over trillions of dollars of wealth while turning over vast power to bureaucrats. Sorry, but no.
Just like any model. When it makes accurate predictions about the actual systems in question ahead of time with minimum effort.
If your model doesn’t match reality, it isn’t reality that’s wrong.
This is a pretty good summary. The problem I see with climate models is they’re attempting to predict very low frequency phenomena, so it’s tough to tell if it’s making accurate predictions.
It’s worse than that – most of the predictions made in their first report have already been falsified.
There are a couple of websites http://www.windy.com and http://www.ventusky.com where you can see nice animations of the weather models in action. A long time ago I was an aviation weather forecaster for a few years and sure wish we had had those then. It is interesting the watch the 7 day prediction and day by day see how the prediction changes for that day as it gets closer.
Paul wrote: “Is there any numerical-methods intuition as to why a large-scale and chaotic system that is not predictable on a short time scale becomes predictable on a much longer time scale?”
EXACTLY THAT. I’d like a formal proof of that please before ANY further climate predictions based on these models are made.
Pug wrote: “High frequency random noise superimposed on meaningful low frequency dynamics?”
How do you know the low frequency stuff isn’t random noise also?
Sure in the longer term there is the physical constraint of the (slightly variable and more so in the ultraviolet) solar energy input but there are physical constraints on the short term weather stuff too.
In any case the models only reflect the assumptions built in to them and as Freeman Dyson has pointed out, ignore biology totally. Biology of course includes things like vegetation cover which changes albedo and far more subtle things like complex iodine compounds released by Krill in the oceans around Antarctica which affect cloud nucleation. In the long term these are important as shown by the current greening of the planet. Can’t figure why that’s a bad thing.
I’m living in Australia, the world’s crash test dummy for “renewable” energy and boy, are we ever dummies. Hundreds of years worth of coal and 40% of the easily recoverable uranium and we have no nukes and are shutting down and blowing up coal plants and putting up with power cuts in major cities because demand exceeds generation capacity. Our renewable sources produced 4% of load during the hot weather the other day.
I don’t know for certain that it’s not noise (I’m not also prescribing radical changes to world economies based on these models), but chemical and geological records indicate multiple slow moving phenomena like ice ages.
The classic test for any physical system model that requires tuning from observation to achieve accuracy is: once the model has been tuned to one set of data, can it reproduce results from another set of data to which it was not tuned?
An example that I’m intimately familiar with: flight dynamics models of aircraft. I have derived a set of differential equations that describe the airplane’s dynamics, given a certain set of assumptions; said equations have parameters that I know a priori – wind tunnel data, structural mode frequencies and shapes, actuator dynamics – but there are certainly errors in my knowledge of the parameters, and there may be errors in the assumptions I used in deriving the equations of motion – e.g., I may have assumed that the aircraft structural modes are not significantly coupled with the aircraft rigid body degrees of freedom. When I go test the aircraft, I compare my model to the truth of the measurements – using a subset of the available data – and adjust the parameters and, if needed, the assumptions that went into my equations of motion, until the model matches the flight results. I then use the updated model to see if I can reproduce the data that I didn’t use in my model-updating process; if I can, then my model has predictive power and is useful for further analyses. If not, then I need to work on my model some more.
This process needs to be applied to the climate models. Take all of the data up through, say, 1960, and adjust a given physical (not correlational) model until it matches. And not just global temperature anomaly either. Going back to my airplane example, if all it predicts accurately is total fuel consumption, it’s not useful; I need it to predict accurately everything from takeoff performance to inner loop stability margins. Similarly for a candidate climate model, it should predict broad trends across each major landmass, ocean surface and depth temperatures across the globe, etc. Now take that model, and run it – with NO additional tweaking – from 1960 to the present day, and see how it does. If the model predicts significant results don’t match reality over that interval – it’s not enough to match sometimes, it has to be a good match across the whole time interval – then it may have predictive power over the next 40+ years and it may behoove us to pay attention to it. If not, it is by definition not a useful model and it is, in the immortal words of John Nance Gardner, “not worth a bucket of warm spit.”
I think there was a different bodily fluid in the bucket in Garner’s original formulation.
Of course another issue is that if you do this process “many times” until the second data set matches as well, you have really just tuned to both data sets and your model still isn’t worth anything.
For the model to be worth something, it must work on the second data set first time. Not on the hundredth try.
That’s basically what we’re dealing with. The climate model projections are essentially extrapolations of an elaborate curve fit. With a rich enough functional basis, one can fit any model to a historical baseline with arbitrary precision. But, that alone does not establish the model as valid. It is necessary, but not sufficient.
Looking over the long term climate record, Earth’s climate exhibits the character of a strange attractor (https://en.wikipedia.org/wiki/Attractor#Strange_attractor) with 2 major modes: A warm mode that we are currently in, and a cool mode. The long term climate prediction challenge is predicting when the mode will change. Shorter term the variations are of smaller magnitude.
I’ve put an embarrassingly large amount of time into studying and thinking about climate models. I certainly resonate with cthulhu’s post, above, especially with: “The classic test for any physical system model that requires tuning from observation to achieve accuracy is: once the model has been tuned to one set of data, can it reproduce results from another set of data to which it was not tuned?”
That test can be used to screen climate models without actually doing the experiment. If a climate model is based in any part on CFD solutions of the Navier-Stokes equations, the answer is a simple and resounding “no.”
When even the best CFD software is used to analyze flow involving turbulence (and turbulence dominates Earth’s hydrodynamics), it must be tuned to the boundary conditions by test. Anything from a new pipe elbow shape to a new aircraft outer moldline constitute new boundary conditions. Tuning CFD to the new boundary conditions (and flow initial conditions) will allow one to make predictions of the results of minor variations in flow initial conditions. But change the boundary conditions, and the tuned CFD won’t produce results even close to what one gets in test.
There are fundamental problems with the Navier-Stokes equations, IMHO, but the universally recognized one is turbulent closure. It isn’t mathematically possible with strictly physical constraints. The constraints used to force closure involve various non-physical mathematical equations which match the number of equations to the number of unknowns plus adjustable parameters. The latter are determined by experiment, and fit only one set of circumstances. It is literally a process of curve-fitting a set of data. Interpolation can be done accurately, but extrapolation puts one in unknown territory.
In short, N-S based climate models are not, and can never be, useful for projecting climate.
Pug,
Ice ages. That. They have exactly zero to do with any human influence as did the temperature variations during the current interglacial before the last couple of hundred years for sure.
I tear my hair out when the media drivel comes on TV. Generally takes less than 2 minutes of watching TV news. Maybe I’m getting old and crotchetty.
Does the USA take climate refugees? Not because of climate change but because of governments becoming totalitarian and oppressive in their “solutions” to the non problem while completely wrecking a country’s economy.
Mike,
I don’t think I did a good job of making my first point clear. I’m aware that ice ages had nothing to do with human intervention – I was listing their presence as an example of low frequency climate dynamics. I think you and I are more or less on the same page.
It is difficult to predict high frequency dynamics because they are so sensitive to modeling and measurement error. E.g., a half microsecond delay in a one microsecond cycle completely inverts the phase.
However, it is also difficult to predict low frequency dynamics because one needs very long timelines to observe them. One can fit a low frequency model that appears to track results over a finite time. But, then one generally finds there is a point at which it starts to diverge, as more of the long term process reveals itself.
When are they useful? When they support the political ambitions of left wing political parties.